Volume-weighted bid-ask spread change — BTC feature definition + Q1 2026 backtest record

Volume-weighted bid-ask spread change feature for BTC

This article documents the volume-weighted bid-ask spread change feature used as an input signal in our BTC research stack. The feature is defined deterministically from order-book and trade data; the article reports its definition, the historical-window backtest record, and the operational caveats around using it.

Feature definition

The feature `bid_ask_spread_change_volume_weighted` is computed from Binance BTC spot order-book snapshots (top five levels per side) and the per-bar trade tape. Inputs are: bid and ask prices at each level, level volumes, and per-trade quantity and price.

Step 1 — Volume-weighted bid and ask prices

For each bar, the volume-weighted ask price is the sum over the top five ask levels of (level price × level volume) divided by the total ask-side volume across those five levels. The volume-weighted bid price is computed symmetrically on the bid side.

Step 2 — Per-bar weighted spread and weighted mid price

The weighted spread is the difference between the volume-weighted ask and the volume-weighted bid. The weighted mid is the average of the two. Both are scalars per bar.

Step 3 — Trade-volume integration

For each trade in the bar, the trade quantity is multiplied by the trade price's deviation from the weighted mid. The sum of those products over the bar is the feature output. The sign carries the direction of trade flow at off-mid prices; the magnitude carries the volume scale.

Backtest record (BTC spot, Q1 2026)

The feature was backtested as a single-signal strategy on BTC spot using the Q1 2026 historical window. Execution conditions in the backtest included a round-trip transaction cost of 0.004% and per-trade slippage of 0.0002%. The metrics below are read directly from the backtest record.

Limitations and operational caveats

The feature definition uses only the top five levels of the order book; deeper book changes are not captured. Trade integration in Step 3 uses the per-trade tape as reported by the venue; reporting latency and out-of-band fills are not modelled.

The backtest uses a single historical window (Q1 2026) and a single venue (Binance BTC spot). Walk-forward verification on additional cohorts is the appropriate next-step check before treating this single-signal strategy as a candidate for live deployment.

What this article does not claim

This article does not assert anything about market participant intent, order-book intent, or future price direction. The feature is a numeric input; its predictive value, if any, is established by the backtest record above on the stated window, and only by that record.

Sources